The core machine learning engine in Citeline Study Feasibility is based on gradient boosted decision trees that are trained on Informa’s data and proprietary engineered features. Predictions are made stable by reducing the impact of variability and local overfitting through multiple regression models.
Users can see what trial design elements are having a positive or negative impact on predictions and refine their plans accordingly. The end result is a single platform in which feasibility scenarios can be instantly modeled, optimized, compared, and shared with explain-ability and transparency to an 80% confidence interval.
For each scenario, Citeline Study Feasibility provides a probability of enrollment success. This describes the probability that the trial will successfully enroll the target patient accrual, within the target enrollment duration, given the inputted parameters. Reflecting the uncertainty around probabilities, the resulting predicted enrollment duration can also be viewed within a 20–80% confidence interval. These probabilities are derived from Informa's proprietary enrollment prediction algorithm, which has demonstrated clear superiority over conventional prediction methods.
Clinical trials are notoriously subject to external factors that lead to variable enrollment rates. Designing a trial in such a way that minimizes the degree of variability is equally as important as targeting an overall reduction in cycle times - something that Citeline Study Feasibility makes great strides toward. In this way, sponsors can balance the allocation of resources more efficiently both for an individual trial and across a wider R&D program.